GAMENet: Graph Augmented MEmory Networks for Recommending Medication Combination
Authors: Junyuan Shang, Cao Xiao, Tengfei Ma, Hongyan Li, Jimeng Sun1126-1133
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the effectiveness and safety of GAMENet by comparing with several state-of-the-art methods on real EHR data. GAMENet outperformed all baselines in all effectiveness measures, and also achieved 3.60% DDI rate reduction from existing EHR data. |
| Researcher Affiliation | Collaboration | Junyuan Shang,1,2 Cao Xiao,3 Tengfei Ma,3 Hongyan Li,1,2 Jimeng Sun4 1 Key Laboratory of Machine Perception, Ministry of Education, Beijing, China 2 School of EECS, Peking University, Beijing, China 3 IBM Research 4 Georgia Institute of Technology |
| Pseudocode | Yes | Algorithm 1 Training GAMENet |
| Open Source Code | Yes | We evaluate GAMENet1 model by comparing against other baselines on recommendation accuracy and successful avoidance of DDI. All methods are implemented in Py Torch (Paszke et al. 2017) and trained on an Ubuntu 16.04 with 8GB memory and Nvidia 1080 GPU. 1https://github.com/sjy1203/GAMENet |
| Open Datasets | Yes | Data Source We used EHR data from MIMIC-III (Johnson et al. 2016). Here we select a cohort where patients have more than one visit. In practice, if we use all the drug codes in an EMR record, the medication set can be very large, each day in hospital, the doctor can prescribe several types of medications for the patient. Hence, we choose the set of medications prescribed by doctors during the first 24-hour as the first 24-hour is often the most critical time for patients to obtain correct treatment quickly. In addition, we used DDI knowledge from TWOSIDES dataset (Tatonetti et al. 2012a). In this work, we keep the Top-40 severity DDI types and transform the drug coding from NDC to ATC Third Level for integrating with MIMIC-III. The statistics of the datasets are summarized in Table 2. |
| Dataset Splits | Yes | Evaluation Strategies We randomly divide the dataset into training, validation and testing set in a 2/3 : 1/6 : 1/6 ratio. |
| Hardware Specification | Yes | All methods are implemented in Py Torch (Paszke et al. 2017) and trained on an Ubuntu 16.04 with 8GB memory and Nvidia 1080 GPU. |
| Software Dependencies | No | The paper mentions 'Py Torch' but does not specify its version number or any other software dependencies with their versions. |
| Experiment Setup | Yes | For LR, we use the grid search technique over typical range of hyper-parameter to search the best hyperparameter values which result in L2 norm penalty with weight as 1.1. For our methods, the hyperparameters are adjusted on evaluation set which result in expected DDI Rate s as 0.05, initial temperature Temp as 0.5, weight decay ϵ as 0.85 and mixture weights π = [0.9, 0.1]. For all deep learning based methods, we choose a gated recurrent unit (GRU) (Cho et al. 2014) as the implementation of RNN and utilize dropout (Srivastava et al. 2014) with probability of an element to be zeroed as 0.4 on the output of embeddings. The embedding size and hidden layer size for GRU is set as 64 and 64 respectively, word and memory size for DMNC model is 64 and 16 which is the same as (Le, Tran, and Venkatesh 2018). Training is done through Adam (Kingma and Ba 2014) at learning rate 0.0002. We fix the best model on evaluation set within 40 epochs and report the performance in test set. |